Video-based heart rate estimation based on the PPG technique is a remote optical technique allowing to determine the heart rate and breath rate through the intensity or motion variations. This paper proposes a new monitoring method for simultaneous estimation of heart and breathing rates using an active 3D system with structured light.
Active 3D imaging systems capture information about 3D object shapes using artificial illumination. An active 3D imaging system can be realized using different techniques: time-of-flight, triangulation, motion flow, interferometry. In this work, we investigate triangulation-based systems.
Heart and respiration rates are signals analyzed in proposed systems as crucial vital signs from a medical point of view. Furthermore, the type of respiration is also analyzed with separation between abdominal and torso type. The quality of breathing signals received from various optical systems is also compared. This system is based on different approaches to 3D imaging and includes random light pattern, fridge light pattern and active stereo methods. In the light pattern systems, the movements of the chest and abdominal are associated with breathing and are observed by the video camera as the light pattern deformation. The system based on active stereo method provides the depth map by which the breath signal is extracted. Further analysis of received signals is based on Principal Component Analysis (PCA).
The proposed remote optical technique extracts the blood pulse from an image captured by a vision system. Patient’s face is detected with use of dedicated convolutional neural network for robust face alignment. Based on found features points, a few regions of interest are selected, and then their results are compared. Mean pixel values for three color channels (red, green, blue) are calculated. Active 3D imaging technique enables stabilization and quality improvement of the captured image. Not correlated groups of pixels are eliminated. Then cluster analysis is used on the grounds of multi-features. Measurement data are detrended by empirical mode decomposition EEMD. Subsequently, ICA was applied to recover source signals from the observations. Maximum frequency pick in the FFT spectrum determines HR.
The performance of the proposed system was tested through a series of experiments. Vital signs of 10 persons with reference signals have been detected. Both, breath and heart rates estimates were found to be accurate. The proposed system can be used as a component of more complex medical systems where information about the condition of the examined patient is crucial or important.